Hallvard B Furuseth wrote:
Howard Chu writes:
Of course, back-hdb too. Aside from subtree rename (which back-hdb supports and back-bdb does not) the two backends are functionally identical; I've gotten tired of writing "back-bdb/back-hdb" all the time.
The differences between hdb and bdb look small enough that they can easily be merged, unless they will diverge more in the future or there are arrays of EntryInfo somewhere.
The effort required to do the merge, however trivial it may be (probably not) isn't worth it at the moment, since it offers no gain in functionality and there's currently no problem with maintainability. There's plenty of other things actually worth worrying about.
If you want to spend time changing internal structure definitions, what would actually be useful is doing a survey of structure field alignments, looking for wasted pad bytes and such, particularly in the context of processor cache line alignment. I know we've got a lot of structures that are perfectly well aligned on a 32 bit architecture but are grossly padded on 64 bit, and I suspect we have a lot of cache alignment issues impacting our multiprocessor performance. But again, this is a topic for the -devel list, not the -software list.